Hi, I'm Kalpana!

Nice to meet you!

A little about me - I am an applied researcher who is quite settled in the causal aspect of all problems and I apply the same to deep learning as well. I usually focus on ‘why’ predictions and prescriptions turn out the way they do, and more importantly, if I can break the explanation part midway and organically change and apply to derive a better result.

Much like - if I know

  1. (A and B) and not(Z) → C
  2. (A or D) → E (imagine this to be a causal mapping from a system with probabilities that I haven’t shown you)
  • then I know that - (E and B) and not(Z) → C or {F: a product better or worse than C}.

And while this might seem like uncertainty to some, there are quite a few things I learn from it -

  1. I learn of a new product F - in ligand design, this is great, since we need new out-of-sample candidates.
  2. E might be easier and less expensive to use than A.
  3. Even if I have an uncertain proportion of C and F in my product, I might be able to discover a single transform (a beautiful application of operator theory) that maps F → C and C → C and maintains my model as such but with better efficiency.

Now if I didn’t have those two relations to begin with and just a direct predictive system, all I’d have is A, B, D, Z go into a network and come out with a soft unreasoned estimate as E and C sometimes. And forget the discovery of F because combinations {B,E, not(Z)} may not even exist on data.

This is what I focus on. Add probabilities and all the domain troubles that come with it - and that is my day-to-day. :slight_smile:

I am currently working on getting networks and encoders to expose these kinds of relations as well. So this community has been really helpful!

I also work on operations research problems as and when they come by and I am a hobbyist when it comes to reading across different sciences - which is why I don’t tie myself to one domain, but try to take the immense domain complexity and nuances as they come up in my life…

I’m looking forward to learning more here.
Feel free to connect with me on LinkedIn as well - https://www.linkedin.com/in/kalpana-baheti-b25055162/.


1 Like

So happy to have you here, Kalpana!

Kalpana is hands down one of the most insightful, philosophical yet practical, and knowledgeable people I know in the deep learning space.

It’s great to have you here and I’m looking forward to your contribution…and hopefully an AMA session in the near future.